|Publication number||US7437338 B1|
|Application number||US 11/386,005|
|Publication date||Oct 14, 2008|
|Filing date||Mar 21, 2006|
|Priority date||Mar 21, 2006|
|Publication number||11386005, 386005, US 7437338 B1, US 7437338B1, US-B1-7437338, US7437338 B1, US7437338B1|
|Inventors||George H. Forman, Henri J. Suermondt, Evan R. Kirshenbaum|
|Original Assignee||Hewlett-Packard Development Company, L.P.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (25), Non-Patent Citations (14), Referenced by (23), Classifications (7), Legal Events (5)|
|External Links: USPTO, USPTO Assignment, Espacenet|
Whenever an organization (such as a business organization, educational organization, or government organization) is involved in providing services to customers or other users, various issues, such as problems, tend to arise. For example, whenever a new product is introduced into a market, some users will encounter difficulties due to actual defects of the product or due to users being unable to figure out how to properly use the product. In this example, two types of problems exist: a defect in the product; and lack of information provided to the user regarding use of the product. Typically, several problems can be associated with products or services provided by an organization.
To address problems associated with products or services, an organization may identify one or more of the problems that the organization deems to be the most important, and allocate resources to discover the underlying causes and come up with solutions for the one or more problems. As examples, the fixes that can be implemented can take the form of redesigning a product, implementing a fix of a current product, introducing new documentation or modifying documentation for customers, providing on-line support documentation, and so forth.
Conventional mechanisms and techniques typically do not enable efficient analysis of problems. Usually, the analysis of a problem is performed manually to determine whether the problem should be addressed. Such manual analysis can be time consuming and prone to error.
Some embodiments of the invention are described with reference to the following figures:
Alternatively, the call agent stations 10 can be replaced with self-help servers, such as self-help servers that provide web pages accessible by customers seeking help with respect to various issues. The self-help servers are able to prompt such customers to enter information regarding the issue(s) they are facing. The information entered by customers is collected and communicated to the analysis server 14 for storing in the one or more data sets 22.
The one or more data sets 22 are stored in a storage 20. The storage 20 can be persistent storage (such as magnetic or optical disk drives or non-volatile semiconductor memory devices), volatile memory (such as dynamic random access memories or static random access memories), or other types of storage devices. The storage 20 is connected to one or plural central processing units (CPUs) 18 in the analysis server 14. Alternatively, the one or more data sets 22 are stored on a database system separate from the analysis server 14.
The data collector 15 is executable on the one or plural CPU(s) 18. Also, as depicted in
In the customer support context, a “case” represents an instance of an interaction between a customer (e.g., a user, patron, subscriber, visitor, member, employee, participant, constituent, supplier, partner, etc.) and an organization (e.g., company, manufacturer, store, provider, employer, representative, etc.) that is indicative of satisfaction or dissatisfaction with something at least partially under control of the organization or another party the organization represents, A “category” (e.g., problem, issue, concern, etc.) represents an underlying reason for the interaction (such as satisfaction or dissatisfaction that led to the case), Such categories can reflect problems associated with a product or service (e.g., defects in product design or manufacture, hardware problems, software problems, hard disk problems, battery problems, and so forth), difficulties in understanding how to correctly use a product or service, difficulty obtaining a product or service, difficulty obtaining information about a product or service, concerns about the value of a product or service, desires for features lacking in a product or service, poor experience interacting with the organization, and so forth. Other organization/customer relationships (e.g., employer/employee, government/constituent) will have similar sets of categories reflecting the types of concerns the customers have and the types of control the entities have. In other environments, other types of categories are employed.
Although described in the context of a customer support environment that includes call agents receiving calls at call agent stations, other embodiments of the invention are applicable to other types of systems for other types of organizations (e.g., educational organizations, government organizations, business organizations, financial organizations, and so forth). Also, another embodiment is applicable to customers interacting with the organization directly by entering information through a website or through an automated telephone entering system.
In accordance with some embodiments, the analysis tool 16 in the analysis server 14 is an interactive analysis tool that allows a user to interact with the analysis tool 16 for the purpose of identifying categories relevant for the cases contained in the data set(s) 22. The analysis tool 16 also enables the identification of a trend for one or more identified categories. Trend analysis allows an organization or user to identify categories associated with quantification measures (e.g., total call volume, support cost, etc.) that are increasing or decreasing due to some event (e.g., new product or service release, occurrence of a product defect, holiday, third party software or hardware releases, and so forth). A “quantification measure” includes (1) a quantity of cases in each category; or (2) an aggregate (e.g., sum, average, maximum, minimum, etc.) of a data field (e.g., cost of parts or time to resolve the case) associated with each of the cases in the respective category. The term “quantification measure” is used interchangeably with the term “metric,” and refers to any measure that is aggregated from values representing any aspect of cases that are being analyzed. A “trend” of a category refers to the increasing or decreasing tendency or direction of a given quantification measure associated with cases of the category over some criterion (e.g., time, location, altitude, etc.). Trend can also refer to the given quantification measure staying stable (in other words, not increasing or decreasing). As used here, a trend need not be monotonic (strictly increasing or decreasing). The trend may be a general statistical increase or decrease represented, for example, by an upward or downward sloping moving average.
By analyzing the trend of a category, or plural categories, the organization or user can determine whether issue(s) associated with the category(ies) is (are) worsening or improving (or more generally, changing), and whether some action should be taken with respect to the category (or categories) to reduce the quantification measure(s) associated with the category (or categories). For example, an increase in technical support calls about battery problems can be caused by a problem associated with the supplier of the battery. The increase in the number of technical support calls on a specific issue represents an upward trend that is an indication that the problem is worsening. Worsening or improving trends of an issue can be indicated by either increasing or decreasing quantification measure(s), or alternatively, by changes in other parameters of a model of the quantification measure(s) over time such as changes in periodicity or seasonality. The analysis of the trend can also just focus on a particular time window (such as between a starting time or date and an ending time or date).
Tracking a trend of a category (or plural categories) also allows for an organization or user to determine whether an action taken with respect to a category has resulted in a decrease (or other change) in the measured quantification measure(s) associated with the category. For example, if the battery problem has been resolved by asking the supplier to implement a fix, then the technical support call volume can be monitored to determine whether the call volume is trending downwardly. Another quantification measure that can be monitored is frequency of calls with respect to a category (e.g., frequency of calls pertaining to the battery problem). The frequency of calls can be measured with respect to total number of calls received for all categories, or total number of products sold. The trend analysis can thus provide a return-on-investment indication of whether an action taken with respect to a category has resulted in a reduction of a quantification measure being monitored for the category.
The interactive analysis tool 16 provides an automated mechanism for performing the trend analysis. Also, the output provided by the analysis tool 16 with respect to the trend for a particular category (or categories) can be in an easy-to-read format (e.g., graph, table, etc.) that allows a user (or users) to quickly determine the trend of the category (or categories). Information pertaining to the trends of multiple categories can be presented side-by-side in the output so that a comparison can be made of the trends of multiple categories.
The analysis tool 16 also performs detection of whether a “systematic” change has actually occurred with respect to a given category. Note that as part of a normal business process, a quantification measure (e.g., total call volume, call frequency, etc.), tends to fluctuate due to various normal business factors (seasonal effects, business trends, and so forth). The fluctuation of the quantification measure during normal operation does not necessarily indicate that a systematic change has occurred with respect to a category that warrants allocation of resources to address the issue associated with the category. Rather, a systematic change results from a change that is caused by an underlying change in a product, service, or other aspect of an organization, rather than from changes due to normal operation of the system (e.g., seasonal fluctuations, business trend fluctuations, a change in the number of units sold or in the number of units under contract such as warranty, etc.). Examples of underlying change that can cause a systematic change include product hardware or software defects, poor documentation, and so forth.
In accordance with some embodiments, the analysis tool 16 includes a change detector to detect for a systematic change in the category based on the observed trend of a quantification measure (or plural quantification measures) associated with the category. The change detector in the analysis tool 16 allows for even more efficient and accurate determination of whether a trend of a particular category is actually increasing (such as in response to occurrence of a product/service defect) or decreasing (such as in response to a fix implemented with respect to a product/service defect). The change detector is able to identify the point at which (or close to which) a systematic change has occurred so that a user is better able to distinguish between a change due to normal operation or a systematic change caused by an issue that should be addressed.
The analysis tool 16 also includes a search-and-confirm mechanism according to some embodiments that enables the creation of training cases, based on user input (described further below), for the purpose of training a categorizer or plural categorizers in the analysis tool 16. Note that the cases stored in the data set(s) 22 include unlabeled cases (cases that are not initially identified with any particular category or categories) as well as labeled cases (cases that have been labeled as belonging to a category or plural categories).
In accordance with some embodiments, another module in the analysis tool 16 is a trend detector 140 that is executable to analyze quantification measures computed with respect to a category (or plural categories). The trend detector 140 is able to provide an output of the trend detected for a category (or plural categories). An example output is in the form of the graphs depicted in the chart of
In accordance with some embodiments, the trend detector 140 defines multiple bins based on one or more criteria. As examples, the one or more criteria include a time criterion, a product type criterion, a product number criterion, a product age criterion, a size criterion, a speed criterion, a cost criterion, a warranty type criterion, a customer type criterion, a customer interaction type criterion, an altitude criterion, a temperature criterion, a humidity criterion, another atmospheric condition (e.g., pressure) criterion, a location criterion, a usage criterion (e.g., number of pages printed, mileage, number of landings), a customer income level criterion, a purchase price criterion, or some other criterion. A “bin” refers to any grouping or collection of cases, where the grouping or collection is based on the specified criterion. Note that a particular case can be assigned to a single bin, to no bins, or to multiple bins. For example, the bins can correspond to instances of a sliding time window of a particular length (e.g., one week), which slides one day (or other unit) at a time—in this scenario, a case that occurs on a particular day can belong to seven bins. The bins can be ordered (e.g., ordered according to time, location, altitude, product number, and so forth). For each bin, a quantification measure (or multiple quantification measures) is computed for cases of each category. Note that in each bin, the quantification measure for one category is computed independently of the quantification measure for another category. In one example, a quantification measure can be the number of calls in each bin (e.g., a time window) for a given category (e.g., battery problem). A trend is then determined based on the quantification measures in multiple ordered bins. The trend is displayed as an output, such as in the chart of
If the bins include time windows, then the cases (in a defined larger time period such as over the past quarter, year, or multiple years) are discretized or divided into the time windows. Each case can be associated with a time data field (e.g., time and date that a case was created or modified). The value of the time data field is used to divide the cases into the plural bins. Quantification measures for cases in respective categories are computed in each time window. For a given category, the quantification measures over multiple time windows can then be used to derive an output that depicts, visually, the trend of cases in that category over time.
The trend detector 140 also includes a change detector 142 to detect, based on the quantification measures, whether a systematic change has occurred. As discussed above, a systematic change is indicative of an underlying change (e.g., product or service defect, fix implemented for a defect, etc.) in a product, service, or some aspect of an organization that causes either an increase or decrease in the corresponding quantification measure(s). In accordance with an embodiment, the change detector 142 implements a cumulative sums (CUSUM) algorithm to detect the systematic change. The CUSUM algorithm is based on calculating cumulative sums of a particular quantification measure over time (to produce time series of quantification measures). The CUSUM algorithm is performed with respect to the time series of cumulative sums to determine when a systematic change has occurred.
Alternatively, instead of using the CUSUM algorithm, other algorithms for detecting change can be employed, such as a generalized likelihood ratio (GLR) algorithm. Other algorithms can be used in other embodiments.
The quantifier 126 in the analysis tool 16 cooperates with one or plural categorizers 116 to perform its quantification task. The categorizer(s) 116 is(are) trained by a training module 114, based on training cases 112 (including positive training sets of cases and negative training sets of cases) created based on a search-and-confirm mechanism.
The search-and-confirm mechanism performs a search-and-confirm procedure to develop the training cases 112. Training a categorizer refers to building or modifying a categorizer based on the training cases, including a positive training set of cases and a negative training set of cases for each category.
In some implementations, concurrently with the search-and-confirm procedure for the purpose of developing categorizer training cases, new categories (such as for problems faced by an organization) can be discovered or existing categories modified. As new categories are added or existing categories modified, training cases are further developed for these new or modified categories to further train the categorizer 116.
The trained categorizer 116 is then used to categorize cases of the data set(s) 22 into one or more categories. In alternative implementations, multiple trained categorizers 116 can be used to categorize cases of the data set(s) 22 into one or more categories. As the categorizer 116 categorizes cases, quantification of the cases in each category is performed by the quantifier 126.
The analysis tool 16 also includes an optional ranking module 106 that is able to identify higher priority categories based on detected trends. For example, a category can be ranked higher if a trend indicates any of the following: a strong trend upwardly or downwardly of a quantification measure associated with the category (e.g., categories associated with trends changing at a more rapid rate can be ranked higher); a sharp change in the trend; and extrapolation of the trend into subsequent ordered bins (e.g., into the future) indicates increasing magnitude of the quantification measure (e.g., future predicted values of the quantification measure based on the trend indicates an increasing problem).
Providing information to identify higher priority categories enables ranking of the categories so that an organization can properly allocate resources to more important respective categories. For example, the ranking can enable an organization to identify the top ten (or top predetermined number) problems that should be given more attention. The information identifying the higher priority categories is provided for a predetermined frame of reference, such as a predetermined time period, product line, geographic region, and so forth. In other implementations, the ranking module 106 can be omitted. Ranking the categories by the ranking module 106 enables presentation of a prioritized list of categories to a user for the user to select one of the issues from this prioritized list for the purpose of identifying a fix or other event with respect to the selected issue and for measuring an impact of the fix or other event.
Although the various modules depicted in
The identification of categories is performed interactively by a user of the analysis tool 16 as the user uses the search-and-confirm mechanism of the tool 16. The search-and-confirm mechanism includes a search engine 102 and confirmation module 122. The search engine 102 enables a user to submit a search and to display a list of cases matching the search criterion or criteria. With the confirmation module 122, the user is able to confirm or disconfirm whether each of the displayed cases belongs to a category (or plural categories).
The search engine 102 is able to receive a query from a user through a user interface 118, and to issue the query to identify cases from the data set(s) 22. The search engine 102 displays information regarding identified cases from among the unlabeled cases 110 that match the query. The displayed information regarding the identified cases is presented in the user interface 118. The user interface 118 can be a graphical user interface, according to an example implementation.
The information displayed in the user interface 118 by the search engine 102 in response to the query includes information regarding a subset of the unlabeled cases 110 that match search criterion(ia) (in the form of search terms or patterns) in the query. A case is said to “match” a query if any information associated with the case satisfies some criterion, such as search term, in the query. A “term” specified by a query refers to any string, substring, regular expression, glob expression, non-textual object (e.g., audio object, video object, etc.), or any other term. A “glob expression” is an expression containing an operator indicating presence of zero or more characters (e.g., *), an arbitrary character (e.g., ?), a range of characters, or a range of strings. A case matches a search term in the query if any information associated with the case satisfies the search term in any specified manner (in other words, equality between the case information and the search term is not required, since the query can specify other forms of relationships between the case information and search term). Not all cases that match need to be used. In some scenarios, a query is matched to only certain field(s) of a case, such as a case description entered by a call agent or user. The user interface 118 displays a summary of each of the matching cases to provide a user with information regarding each case. The process of specifying a query and viewing results of matching cases is referred to as a “scooping” process. Following the scooping process, a “confirming” process is performed in which a user is able to confirm whether or not each of the matching cases belongs to a particular category (by selecting or deselecting displayed fields or other indicators).
User confirmation (or disconfirmation) is monitored by the confirmation module 122. Not all displayed cases need to be confirmed or disconfirmed. For cases that have been correctly matched to a category, such cases are added to a positive training set of cases. On the other hand, for cases that have been incorrectly matched, the confirmation module 122 adds such incorrectly matched cases to a negative training set of cases. The positive and negative training sets, which are part of the training cases 112 stored in the data set 22, are accessed by the training module 114 for training the categorizer 116.
The search engine 102 and confirmation module 122 thus cooperate to develop training cases 112 from cases in the data set 22 (based on user confirmation and disconfirmation), which training cases 112 are used by the training module 114 to train the categorizer 116.
During the searching and confirming, a user can determine that additional categories should be added to a hierarchy 124 of categories, or existing categories in the hierarch 124 modified. Using a category editor 132 in the analysis tool 16, the user can move, add, modify, or even delete categories (represented by the hierarchy 124 of categories stored in a storage 20). In the example hierarchy 124 depicted in
The category editor 132 is responsive to user input at a user interface (UI) 118 (presented in a display monitor 120) to add categories (or subcategories), delete categories (or subcategories), or modify categories (or subcategories). In response to user input to add, delete, or modify categories (or subcategories), the category editor 132 is able to modify the hierarchy 124 of categories.
Note that, initially, there may already be a developed set of categories before the search-and-confirm procedure is started, which existing set of categories can be used as a starting or initial set of categories. In an alternative scenario, such as with a new project, no categories may exist. In this alternative scenario, the user may create one or a few categories as the starting point, or the one or few categories can be created by another technique, described further below.
In one embodiment, the categories in the hierarchy 124 of categories are in a directed acyclic graph (DAG) rather than a tree. In other words, any category in the hierarchy 124 can have not only several children, but also several parents. However, a category cannot be simultaneously an ancestor and a descendant of another category. Subcategories (associated with a particular category) are considered the children of the particular category. In alternative implementations, other structural relationships of categories can be employed.
A manager module 101 in the analysis tool 16 performs overall management operations, such as managing the storing of data (including training cases 112 and hierarchy 124 of categories) in the storage 20 and coordination among the various modules of the analysis tool 16.
As the positive and negative training sets are modified based on the user confirming and disconfirming acts, and based on modification of the hierarchy 124 of categories, the modified positive and negative training sets are propagated through the hierarchy 124 of categories to enable the training module 114 to train the categorizer 116 for the categories.
During development of the categorizer 116, the quantifier 126 is also created by a quantifier creator module 128 in the analysis tool 16. The quantifier 126 can be in any format, such as an Extensible Markup Language (XML) format, C code format, or any other format. In the arrangement of
The quantification performed by the quantifier 126 includes computing, for one or more of the categories in the hierarchy 124 of categories, a quantification measure that represents the number of cases in each category. Alternatively, the quantifier 126 is able to generate another quantification measure, such as a sum or some other aggregate of a data field associated with the cases that belong to each category. As an example, the quantification can represent a measure (e.g., sum) of one or more of the following: the duration of calls; cost of repair; amount of time to resolve the calls; amount of lost revenue; an amount of money to resolve calls; amount of lost revenue due to calls; a degree of customer aggravation; an amount of time spent by customers before initiating a call; an amount of time spent by customers during the call; an amount of time spent diagnosing or resolving cases; an amount of money spent by customers; an amount of money spent interacting with customers; an amount of money spent diagnosing or resolving the cases; and a number of customers who declined to initiate a call; and so forth. The quantification measure can be adjusted based on some secondary factor, such as number of units sold or number of units under warranty. Other examples of aggregation include an average or mean, a standard deviation, a 95% (or other percentage level) confidence interval, a variance, a minimum, a maximum, a median, a mode, a geometric mean, a harmonic mean, a percentile rank, an ordinal statistic, or other statistic of the values in the data field associated with the cases or a value computed based on fitting a model to the values in the data field associated with the cases. The data field of a case can contain a single value or a collection of values, or the data field can be a value or collection of values computed from other values associated with the case. In some embodiments the aggregate may involve a plurality of data fields associated with each case. The quantification measures computed by the quantifier 126 are stored as quantification measures 134 in the storage 20.
Note that the quantification measures computed by the quantifier 126 are estimated measures in light of the fact that the categorization performed by a categorizer 116 is often not absolutely accurate. For example, the accuracy of a trained categorizer can be impacted by the quality of training cases used to train the categorizer. As explained further below, to compensate for inaccuracies of the categorizer 116, calibration of the quantifier 126 is performed based on some indication of the ability of the categorizer 116 to categorize cases in a data set or based on the statistical distribution of scores issued by the categorizer 116 on cases.
The quantifier 126 is able to display the quantification measures it computes either numerically or graphically. Also, the quantifier 126 is able to generate one or more reports 138 that present these quantifications. The reports 138 optionally also provide comparative quantifications, such as providing a comparison of cases in different data sets or in different partitions of one data set.
As discussed above, the analysis tool 16 provides an interactive software package that simplifies the processes of identifying new categories for cases (unlabeled cases 110 and/or labeled cases 111) in the data set(s) 22; developing training cases for training the categorizer; training the categorizer; categorizing cases; quantifying the categorized cases; and providing information regarding a trend of one or more categories.
A general flow of general tasks performed by the analysis tool 16 is discussed in connection with
After classification, the cases are allocated (at 158), such as by sorting or dividing, into multiple ordered bins (e.g. time windows, locations, altitude ranges, etc.). For cases in each category and in each bin, a quantification measure (or quantification measures) is(are) computed (at 160). The analysis tool 16 then provides (at 162) information regarding the trend of the quantification measures in the ordered bins associated with each category being studied. The provided information regarding the trend of each category being studied can be depicted visually, such as in the chart of
The GUI screen 200 includes a first frame 202 that displays the hierarchy of categories (also referred to as “classmap”) that is being built using the analysis tool 16 of
The GUI screen 200 also includes a second frame 204 with a “training” column 206 to display the number of cases that have been identified by a user (either directly or by inference) as being positive cases for a given category. The “Training” column 206 lists the number of cases for each of the “sync,” “screen,” “batt,” and “sdcard” categories. For example, in the “Training” column 206, the “sync” category has 93 cases, the “screen” category has 200 cases, the “batt” category has 2,394 cases, and the “sdcard” category has 47 cases.
Another column in the second frame 204 is an “Estimate” column 208, which displays the estimated number of cases identified by the quantifier 126 as belonging to each of the categories and subcategories. A third “Quantity” column 210 lists an estimated sum of some data field of the cases in each category, such as time spent on cases that belong in the category or cost of handling the cases in the category. The values in the “Quantity” column 210 are also provided by the quantifier 126. As other examples, the “Quantity” column 210 can indicate costs or percentages of volume of cases.
Another optional column in the second frame 204 can indicate the perceived quality of the current categorizer with respect to each particular category. Examples of information pertaining to the measure of the quality of the categorizer 116 that has been trained using training information developed according to some embodiments includes: a true positive rate (the likelihood that an item in a category will be identified by the categorizer to be in the category); a false negative rate (the likelihood that an item in a category will be identified by the categorizer to be not in the category); a true negative rate (the likelihood that an item that is not in a category will be identified by the categorizer to be not in the category); a false positive rate (the likelihood that an item that is not in a category will be identified by the categorizer to be in the category); an accuracy measure (the likelihood that an item will be correctly identified to be or not to be in a category); a recall measure (same as true positive rate); a precision measure (the likelihood that an item identified to be in a category actually is in the category); a bi-normal separation measure (a measure of the separation between the true positive rate and the false positive rate); an information gain measure (a measure of the decrease in entropy due to the categorizer); a lift measure (a measure of an increase in, e.g., response rate if only the cases the categorizer is most confident about are processed); a measure of stability under cross-validation (measure of the likelihood that the categorizer has or has not overfit the training information by learning to recognize individual cases rather than learning generalities that apply to unseen data); a measure of an area under an ROC (receiver operating characteristic) curve (area under a curve that is a plot of true positive rate versus false positive rate for different threshold values for a categorizer); a number of training cases; a percentage of target training size (same as number of training cases, except with the added notion that a user has indicated a desire to see a minimum number of cases for every category); an F-measure (a parameterized combination of precision and recall); a total cost (an expected aggregate cost over analyzed cases considering separate individual costs for the true positives, true negatives, false positives, and false negatives); and an average cost (similar to total cost, except averaged over the number of cases).
A search frame 212 is also displayed in the GUI screen 200. The search frame 212 includes a search area 214 in which a user can enter a query for performing the scooping process. As part of the scooping process, the user-provided query is submitted to the search engine 102 (
The display frame 216 displays a summary (e.g., title) of some or each of the cases identified by the search based on the query entered in the search frame 212. Note that each case is associated with several pieces of information, with the title being one of the pieces. In other implementations, other pieces of information associated with the cases can be displayed. In the example of
If the result cases do not fit in one page, a scroll bar 217 is provided to enable the user to scroll to the remaining cases. Alternatively, a user may specify that the GUI screen 200 displays a set of randomly selected cases that fit within the display frame 216, such that the scroll bar 217 would not have to be used.
In the third column 222 of the display frame 216, a summary of the case, such as the case title, is illustrated. For example, the summary provided can have been previously entered by a customer support representative when answering customer calls. Even though the displayed summaries may contain mis-spellings, grammatical errors, and abbreviations, a user looking at each summary can quickly ascertain whether each respective case is associated with the category in question.
Note that each of the case titles displayed in the third column 222 of the display frame 216 contains the word “charge.” Based on the displayed case title, a user can select or de-select each user-selectable box in the second column 220. In other implementations, other types of summaries can be displayed in the third column 222 to provide information that the user can use to select or de-select boxes in the second column 220. Selection of a box indicates that the user has confirmed that the particular case belongs to the category. On the other hand, when a user de-selects a box in the second column 220, that is an indication that the corresponding case does not belong to the category in question (that is, the user has disconfirmed that the case belongs to the category).
In a different embodiment, instead of displaying just one category, the display frame 216 can display multiple categories with fields that are user-selectable to confirm or disconfirm whether a case belongs to the categories. A confirmed case can be added to a positive training set, while a disconfirmed case can be added to a negative training set. The positive and negative training sets are used to train the categorizer.
As a user labels cases as belonging (or not belonging) to particular categories based on input to the confirmation module 122, training cases 112 (positive and/or negative training cases) are added or modified, such as by the confirmation module 122 or by some other module in the analysis tool. As the training cases 112 are added, the categorizer 116 is trained concurrently. The training of the categorizer 116 as positive training cases or negative training cases are added can be performed in the background (according to some embodiments) so that the training (or retraining) of the categorizer 116 does not interfere with the search and confirm processes used for identifying training cases. The trained categorizer is installed atomically once the training is complete. If the user makes changes to the categories while the categorizer is training, the training can be stopped and restarted with the modified categories. Note that the term “training” refers to either the first training of a categorizer or a retraining of the categorizer.
Also, as categories are added, modified, and/or deleted, the categories displayed in the first frame 202 are changed by the category editor 132 (
As shown in
Examples of the types of information that are received and stored include information such as the time and duration of call; information about the product being called about, replacement parts ordered, compensation offered, or repair people sent; information automatically gathered from a computer of a customer; uniform resource locators (URL(s)) or other resources visited and documents consulted in attempting to solve the problem; linked information on the demographics of the customer (e.g., location, age, gender, technological sophistication, loyalty, generated revenue, etc.); call agent-entered description of the state-of-mind of the customer; description of interaction between call agents and customers; call agent selection of a category or problem based on a call; transcripts or recording of on-line chat sessions or telephone conversations; call agent-written short summaries of a problem; customer e-mail messages; records of pages visited on a web site (such as on-line documents viewed by customer or call agent); outputs from a diagnostic program; data stored in a database (e.g., data relating to a prior support call, information relating to a purchase, and demographic information of a customer); computer files received from a customer; survey data; data received by a monitoring device; and others.
Steps 302 and 304 are repeated as additional information regarding issues are received.
Next, a search-and-confirm procedure (
The analysis tool 16 reads an initial hierarchy of categories (at 310). One way of creating the initial hierarchy categories is based on user input. In one scenario where training cases are to be developed for a new project, a set of categories may not yet exist. In this scenario, a user can create one or a few categories as the starting point. An expert may create the categories based on a preconceived notion of the relevant categories (such as problems) associated with a particular project. Alternatively, the expert may create the one or few categories based on prior experience or common knowledge of the expert. For example, it may be well known that a product has problems with battery life, wireless connectivity, and keys following off. In a second approach, an expert may “eyeball” cases by scooping random samples and seeing which problems jump out at the expert. A user or expert can add, delete, or modify categories using the category editor 126 (
Alternatively, instead of relying on a human user or expert, an automated module in the analysis tool 16, can be used to examine the unlabeled cases 110 and determine based on this examination one or plural possible categories. For example, one technique that can be used by such a module is a cluster-detecting technique based on a clustering algorithm to identify groupings of cases. These groupings of cases identified are reviewed by a user or expert to determine which categories are appropriate to add to an initial set of categories. Note that this initial set of categories provided (at 310) does not have to be highly accurate categories, nor do these categories have to survive the process of developing the categories. The initial set of categories merely provides a starting point.
In a different scenario, there may already be a set of extant categories that can be used as a starting point for further development of training cases.
After one or a few initial categories have been identified (at 319) for the purpose of searching and confirming, the scooping process can begin. To start the scooping process, a user enters a query relating to a category into the search area 214 of the GUI screen 200 (
In response to the query, the search engine 102 identifies (at 322) the matching set of cases and displays (at 324) the identified set of cases in the user interface 118 (
As noted above, the submission of the query, identification of cases matching the query, and the display of the cases is part of the scooping process. Typically, the scooping process has been performed with a hypothesis in the form of “cases that match this query should be training examples for category C.”
After the cases have been displayed by the search engine 102, then the confirmation module 122 can receive (at 326) user confirmation or disconfirmation. For example, some of the cases may have been identified or inferred as being or not being in the category (or categories) in question. In the example of
In an embodiment, the categorizer can determine whether a matching case should be indicated as belonging to a category by computing a confidence indication. The confidence indication is compared to a predefined threshold, and if the confidence indication is greater than the predefined threshold, the categorizer identifies the matching case as belonging to the category.
The user next goes through the displayed cases and either confirms or disconfirms by respectively checking the box (or leaving the box checked) or un-checking the box (or leaving the box un-checked) in the column 220. Note that a case can belong to more than one category, so that a scoop for one category may return cases that have already been labeled as belonging to another category. Note that check boxes constitute one example implementation for confirming or disconfirming that a case belongs to a category. There are numerous other techniques in other implementations, including techniques to check plural boxes at the same time.
For those cases that have been indicated as belong to the category in question (based on user selection of the box in column 220 in
Steps 319-328 are repeated to develop training cases for other categories or to more fully develop training cases for a current category.
As part of the process of confirming or disconfirming cases, the user may realize that a new category should be created. In the example depicted in
The modification of the hierarchy 124 of categories can result in changes of the positive and negative training sets, which changes are propagated (at 334).
Since any added or modified category is based on a user's examination of the summaries of cases listed in response to the query, the added or modified category is likely to be semantically meaningful (the label for the category or subcategory is descriptive of the corresponding problem or other event or item). Also, the set of categories created is not biased by any requirement that all cases have to be labeled or that all cases have to be assigned to a single category.
As discussed, as the user performs confirm/disconfirm and as categories or subcategories are added/modified/deleted, changes occur in the positive and negative training sets, which are propagated (at 340) through the hierarchy 124 of categories, as depicted in
As the training cases 112 are being updated by the scooping and confirming processes described above, the categorizer 116 is trained (at 346) in the background for categories whose positive or training sets have changed. Optionally, the retraining (at 346) can be delayed, by use of a timer involving optional steps 343 and 344. The timer is set or updated (at 343) to expire after a predetermined amount of time. Any change to the training cases will cause the process to loop back to step 340 (which will cause the timer to again be set or updated at 343). After some period of stability (a period during which no changes occur) has been detected, the timer expires (at 344), which allows the categorizer retraining (at 346) to be performed. Checking for stability at 344 avoids the situation where many successive changes in the positive and training sets in a short period of time caused by user confirming/disconfirming or category modification causes the retraining to restart many times. The retraining of the categorizer 116 can occur generally concurrently with the scooping and confirming processes so that the user can continue to perform the scooping and confirming processes even while the categorizer is being trained, subject to the training module 114 (optionally) waiting for expiration of the timer to ensure some period of stability. An output provided by the scooping process, confirming process, and training process described above is a categorizer, or plural categorizers.
The retrained categorizer is invoked to re-categorize or relabel (at 348) the cases. Also, the quality of categorization or labeling performed by the categorizer can be determined (at 350) (by measuring the categorizer), and a measure of this quality can be displayed (at 352). The quality measures of a categorizer are listed above. If the quality measures of the categorizations performed by the trained categorizer indicate that the categorizations are either incorrect or of insufficient confidence (quality measure being less than a predefined threshold), the categorizer can be retrained again.
Steps 340-352 are repeated as positive and/or negative training sets are modified for the various categories.
As depicted in
In one embodiment, to calibrate the quantifier 126 with respect to a category C, the analysis tool 16 determines the fraction (TPR) of the time that the categorizer is correct when presented with a case that should be categorized in a category C (also referred to as the true positive rate), and the fraction (FPR) of the time that the categorizer is wrong when presented with a case that should not be categorized in the category C (also referred to as the false positive rate). Assuming that a data set includes a total of T cases, of which N cases have been categorized in the category C, then the calibration to be performed by the calibrated quantifier 126 on the value N is as follows:
where N′ is the calibrated N value.
Similar fractions representing true positive rates and false positive rates are also maintained for the other categories to enable calibration of the quantity values for other categories. Other techniques for calibrating a quantifier/categorizer are described in U.S. Ser. No. 11/080,098, entitled “A Method of, and System for, Classification Count Adjustment,” filed by George H. Forman et al., on Mar. 14, 2005. In other embodiments, any other technique of calibrating a quantifier or adjusting its output quantification, especially by observing the behavior of its underlying categorizer(s), may also be used.
In yet another calibration technique, a score associated with each of a group of cases is obtained from the categorizer. A statistical distribution is computed based on the scores. The quantification measure is adjusted based on the statistical distribution.
The calibrated quantifier 126 is used to update (at 364) quantification measure(s) for each of the categories. The calibrated quantifier 126 produces calibrated quantification measure(s). The categorizer(s) 116 associated with the quantifier 126 provides information regarding the categorized cases to the quantifier 126 to enable the quantifier 126 to produce the quantification measure(s). Note that the quantification process can proceed concurrently with the search-and-confirm and the training processes discussed above. Thus, the analysis tool 16 may present running estimate(s) of the quantification measure(s) in each category. The quantifier 126 displays (at 366) the updated quantification measures, such as estimated number of cases and quantity estimates (columns 208 and 210 in the frame 204 of
Steps 362-366 are repeated to continually update quantification measure(s) as more cases are received and the categorizer 116 is retrained.
As discussed above, the trend detector 140 in the analysis tool 100 is able to detect a trend of at least some of the categories (e.g., customer support issues). As depicted in
If the criterion is a time criterion, then the bins are time points or time windows, where the trend analysis is performed for predefined time points or time periods (e.g., weekly periods, monthly periods, quarterly periods, etc.). Time windows can be overlapping or non-overlapping. One type of overlapping time windows is sliding time windows. As an example, a first sliding window can be defined as including Day 1 through Day 7, a second sliding window can be defined as including Day 2 through Day 8, a third sliding window can be defined as including Day 3 through Day 9, and so forth. In the sliding window example, each sliding window is considered a bin.
If the criterion for defining the bins is product number or product type, then each bin contains cases for a specific product number or product type. Similarly, if the criterion for defining the bins is altitude, then each bin can be defined by a range of altitude levels (e.g., bin 1 includes ground level to 5,000 feet, bin 2 includes 5,001 feet to 10,000 feet, and so forth).
The bins are then ordered (at 412) according to the corresponding criterion. If the criterion is a time criterion, then the ordering of the bins is based on time so that a trend over time of cases in the multiple bins can be detected. Alternatively, if the criterion used is an altitude criterion, then the bins are ordered according to altitude so that a trend at different altitudes (e.g., use of a product when flying versus use of a product at sea level) can be detected.
Cases that are under analysis are then allocated to the bins (at 414) by the trend detector 140. The cases under analysis can be recent cases that have been received in the latest month, quarter, or other time period. Note, however, for the purpose of detecting a trend, the cases under analysis can include cases over longer periods of time (e.g., one or several years). The cases under analysis have been classified into various categories by the trained categorizer (see
For each bin, a quantification measure (or multiple quantification measures) is (are) computed (at 416) by the quantifier 126 for cases in each category. As an example, a quantification measure can be the number of calls in each bin (e.g., a time window). Other quantification measures (including those listed further above) can be used as other forms of metrics in other implementations. Thus, for each bin, multiple quantification measures can be computed for corresponding categories. For example, a first number of calls for a first category (e.g., battery problem) can be computed, a second number of calls for a second category (e.g., display failure) can be computed, and so forth. Alternatively, the trend detector 140 can focus on just one category so that a single quantification measure is computed for each bin.
The trend detector 140 then produces (at 417) an output (for presentation to a user) that contains indications of the quantification measures for each category in the ordered bins (e.g., time series, altitude ranges, product models, etc.). The output enables determination of a trend for each category over the ordered bins. As examples, the output can be a text-based report (e.g., a series of numbers representing the quantification measures for each category), a graph (such as a pie chart, bar chart, stacked bar chart, line chart, scatter plot, bar-and-whiskers plot), or a report according to any other format (e.g., a table).
Note that there are several types of trends: (1) trends that vary between increasing, decreasing, and flat; (2) a steadily increasing or decreasing (steadily changing) trend; and (3) a trend characterized by a sharp increase or decrease (sharp change). A steadily changing trend can either be a gradually increasing or decreasing trend, or a strongly increasing or decreasing trend. A trend according to type (1) usually occurs during normal operation.
However, a trend according to type (2) or (3) usually results from some underlying factor (such as a systematic change). A trend according to type (2) or (3) is usually an indication that something unexpected has occurred (e.g., product defect, third party new software release, etc.). Consequently, detection of a trend according to type (2) or (3) is usually an indication that an organization or user should focus resources to address an issue (or issues) associated with a category (or categories).
As further depicted in
Based on the detected trends associated with several categories, the ranking module 106 can rank (at 419) the categories. For example, a category can be ranked higher if a trend indicates any of the following: a strong trend upwardly or downwardly of a quantification measure associated with the category (e.g., categories associated with trends changing at a more rapid rate can be ranked higher); a sharp change in the trend; and extrapolation of the trend into subsequent ordered bins (e.g., into the future) indicates increasing magnitude of the quantification measure (e.g., future predicted values of the quantification measure based on the trend indicates an increasing problem). More generally, ordering of the categories is based on at least one of: (1) relative rates of change of the trends of respective categories, (2) presence or lack thereof of sharp changes in the trends of respective categories, and (3) relative magnitudes of future quantification measures of respective categories based on the trends.
The ranked categories can be listed in a report (which can include a chart, table, or other output) that can be provided to a user. The content of the report can be used by the user to determine which categories should be allocated resources to address.
As depicted in
The change detector 142 determines (at 424) whether a systematic change has occurred. This determination can be performed by comparing each of the cumulative sums calculated by the cumulative sums algorithm against a threshold. If a cumulative sums crosses a threshold, then that is an indication that a systematic change has occurred. The change-point detection algorithm applied by the change detector 142 also enables the change detector 142 to find a change point, which is the point (e.g. a time point) at which the systematic change occurred. Examples of change-point detection algorithms that can be used by the change detector 142 include those described in U.S. Ser. Nos. 11/119,037, entitled “Detecting Change in Data,” by Jerry Z. Shan; 11/117,989, entitled “Performing Quality Determination of Data,” by Jerry Z. Shan; 11/118,832, entitled “Determining a Time Point Corresponding to a Change in Data Values Based on Fitting With Respect to Plural Aggregate Value Sets.” by Jerry Z. Shan; U.S. Patent Application Publication No. 2005/0096866, entitled “Techniques for Monitoring a Data Stream.” by Jerry Z. Shan et al.; and U.S. Patent Application Publication No. 2006/0173559, entitled “Methods and Systems for a Prediction Model,” by Evan Kirshenbaum et al.
Identifying the change point along with providing the indication that a systematic change has occurred allows an organization to go back to the point at which the failure (or some other issue) occurred to better understand the issue. The indication of the systematic change and identification of the change point can be output in a user interface (such as a graphical user interface) or provided in some other report.
As noted above, tracking a trend of a category (or plural categories) also allows for an organization or user to determine whether an action with respect to an issue associated with the category should be taken. For example, an action can be taken if the trend indicates that the issue associated with the category is worsening. After the action is taken, the steps of
An “action” taken with respect to a category includes a fix or some other action. Examples of actions that can be taken with respect to a category include one or more of: allocating a sum of money (or otherwise defining a budget); allocating a physical resource (e.g., equipment); hiring a person; assigning a task to a person; writing a document (such as a help document); modifying an existing document; identifying a document; altering availability of a document (such as to make the document more widely available such as posting the document on a web site); altering an organization of a web site; modifying a design of a product; modifying a packaging of a product; modifying a manufacturing process for a product; creating a software program; modifying a software program; creating a patch for a software program; contacting a customer, vendor, supplier, employee, or partner; modifying a marketing campaign; changing response time of service providers; training service personnel; discontinuing efforts that are no longer required; changing the process of writing and delivery of software programs; taking actions with reference to seasonal fluctuations; provide reports to customers regarding how issues are being monitored and addressed; and other actions.
Examples of other events that may affect a particular issue associated with a category include external events (associated with third parties) such as a new version of an operating system, upgrades of operating systems, new knowledge documentation, new hardware platforms, or any other event that can directly or indirectly affect cases associated with a particular issue.
For example, if a problem has been resolved by implementing a product fix, then the technical support call volume on the specific issue(s) can be monitored to determine whether the call volume exhibits a downward change (downward trend). The change detector 142 can be invoked to determine whether a systematic change has occurred. The downward trend along with an indication of change provided by the change detector 142 is an indication that the fix has been successful.
Alternatively, determining whether an action has been effective with respect to a category can be based on a comparison of a first trend (detected prior to the action) with a second trend (detected after the action) determined respect to a particular category.
Instructions of the various software modules described above are loaded for execution on corresponding processors. The processors include microprocessors, microcontrollers, processor modules or subsystems (including one or more microprocessors or microcontrollers), or other control or computing devices. As used here, a “controller” refers to hardware, software, or a combination thereof. A “controller” can refer to a single component or to plural components (whether software or hardware).
Data and instructions (of the software) are stored in respective storage devices, which are implemented as one or more machine-readable storage media. The storage media include different forms of memory including semiconductor memory devices such as dynamic or static random access memories (DRAMs or SRAMs), erasable and programmable read-only memories (EPROMs), electrically erasable and programmable read-only memories (EEPROMs) and flash memories; magnetic disks such as fixed, floppy and removable disks; other magnetic media including tape; and optical media such as compact disks (CDs) or digital video disks (DVDs).
In the foregoing description, numerous details are set forth to provide an understanding of the present invention. However, it will be understood by those skilled in the art that the present invention may be practiced without these details. While the invention has been disclosed with respect to a limited number of embodiments, those skilled in the art will appreciate numerous modifications and variations therefrom. It is intended that the appended claims cover such modifications and variations as fall within the true spirit and scope of the invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5819258||Mar 7, 1997||Oct 6, 1998||Digital Equipment Corporation||Method and apparatus for automatically generating hierarchical categories from large document collections|
|US5930764 *||Aug 23, 1996||Jul 27, 1999||Citibank, N.A.||Sales and marketing support system using a customer information database|
|US5943670||Nov 21, 1997||Aug 24, 1999||International Business Machines Corporation||System and method for categorizing objects in combined categories|
|US6038560||May 21, 1997||Mar 14, 2000||Oracle Corporation||Concept knowledge base search and retrieval system|
|US6170025 *||Aug 11, 1998||Jan 2, 2001||Intel Corporation||Distributed computer system supporting remote interrupts and lock mechanism|
|US6173279 *||Apr 9, 1998||Jan 9, 2001||At&T Corp.||Method of using a natural language interface to retrieve information from one or more data resources|
|US6701333||Jul 17, 2001||Mar 2, 2004||Hewlett-Packard Development Company, L.P.||Method of efficient migration from one categorization hierarchy to another hierarchy|
|US6704905||Dec 28, 2001||Mar 9, 2004||Matsushita Electric Industrial Co., Ltd.||Text classifying parameter generator and a text classifier using the generated parameter|
|US6728690||Nov 23, 1999||Apr 27, 2004||Microsoft Corporation||Classification system trainer employing maximum margin back-propagation with probabilistic outputs|
|US6767213 *||Mar 15, 2002||Jul 27, 2004||Management Research Institute, Inc.||System and method for assessing organizational leadership potential through the use of metacognitive predictors|
|US6823323 *||Apr 26, 2001||Nov 23, 2004||Hewlett-Packard Development Company, L.P.||Automatic classification method and apparatus|
|US6842751||Jul 31, 2000||Jan 11, 2005||International Business Machines Corporation||Methods and apparatus for selecting a data classification model using meta-learning|
|US6944616||Nov 28, 2001||Sep 13, 2005||Pavilion Technologies, Inc.||System and method for historical database training of support vector machines|
|US6973452 *||May 18, 2004||Dec 6, 2005||Netezza Corporation||Limiting scans of loosely ordered and/or grouped relations using nearly ordered maps|
|US7043492||Aug 22, 2002||May 9, 2006||Requisite Technology, Inc.||Automated classification of items using classification mappings|
|US7113957||Mar 26, 2002||Sep 26, 2006||Ncr Corporation||Row hash match scan join using summary contexts for a partitioned database system|
|US7177855||Nov 6, 2003||Feb 13, 2007||Oracle International Corporation||Compile-time optimizations of queries with SQL spreadsheet|
|US20020107712||Dec 12, 2000||Aug 8, 2002||Lam Kathryn K.||Methodology for creating and maintaining a scheme for categorizing electronic communications|
|US20030187809||Mar 29, 2002||Oct 2, 2003||Suermondt Henri Jacques||Automatic hierarchical classification of temporal ordered case log documents for detection of changes|
|US20040064464||Oct 1, 2002||Apr 1, 2004||Forman George Henry||Hierarchical categorization method and system with automatic local selection of classifiers|
|US20050096866||Oct 31, 2003||May 5, 2005||Shan Jerry Z.||Techniques for monitoring a data stream|
|US20060173559||Jan 31, 2005||Aug 3, 2006||Evan Kirshenbaum||Methods and systems for a prediction model|
|US20060248054||Apr 29, 2005||Nov 2, 2006||Hewlett-Packard Development Company, L.P.||Providing training information for training a categorizer|
|US20070185901||Apr 12, 2007||Aug 9, 2007||International Business Machines Corporation||Creating Taxonomies And Training Data For Document Categorization|
|EP0747846A2||May 29, 1996||Dec 11, 1996||AT&T IPM Corp.||Method and apparatus for training a text classifier|
|1||Forman et al., U.S. Appl. No. 11/364,108, entitled "Identifying an Emerging or New Category," filed, Feb. 28, 2006, pp. 1-34, Figs. 1-6.|
|2||G. Forman, U.S. Appl. No. 10/902,924, entitled "System and Method for Category Discovery," filed Jul. 30, 2004, pp. 1-38, Figs. 1-7D.|
|3||G.H. Forman et al., U.S. Appl. No. 11/393,487, entitled "Comparing Distributions of Cases Over Groups of Categories," filed Mar. 30, 2006, pp. 1-39, Figs. 1-7E.|
|4||J.Z. Shan, U.S. Appl. No. 11/117,989, entitled "Performing Quality Determination of Data," filed Apr. 28, 2005, pp. 1-32, Figs. 1-9.|
|5||J.Z. Shan, U.S. Appl. No. 11/118,832, entitled "Determining a Time Point Corresponding to a Change in Data Values Based on Fitting with Respect to Plural Aggregate Value Sets," filed Apr. 29, 2005, pp. 1-26. Figs. 1-8.|
|6||J.Z. Shan, U.S. Appl. No. 11/119,037, entitled "Detecting Change in Data," filed Apr. 29, 2005, pp. 1-26, Figs. 1-6.|
|7||Kirschenbaum et al., U.S. Appl. No. 11/118,786, entitled "Computing a Quantification Measure Associated with Cases in a Category," filed Apr. 29, 2005, pp. 1-35, Figs. 1-3D.|
|8||Kirshenbaum et al., U.S. Appl. No. 11/047,093, entitled "Methods and Systems for a Prediction Model," filed Jan. 31, 2005, pp. 1-29, Figs. 1-6.|
|9||Kirshenbaum et al., U.S. Appl. No. 11/080,098, entitled "A Method of, and System for, Classification Count Adjustment," filed Mar. 14, 2005, pp. 1-31, Figs. 1A-5.|
|10||Kirshenbaum et al., U.S. Appl. No. 11/118,178, entitled "Providing Training Information for Training a Categroizer," filed Apr. 29, 2005, pp. 1-39, Figs. 1-3E.|
|11||Kirshenbaum et al., U.S. Appl. No. 11/172,187, entitled "Effecting Action to Address an Issue Associated with a Category Based on Information that Enables Ranking of Categories," filed Jun. 30, 2005, pp. 1-52, Figs. 1-5H.|
|12||Kirshenbaum et al., U.S. Appl. No. 11/211,979, entitled "Producing a Measure Regarding Cases Associated with an Issue After One or More Events Have Occurred," filed Aug. 25, 2005, pp. 1-51, Figs. 1-5H.|
|13||Office Action dated Oct. 16, 2007 from U.S. Appl. No. 11/118,178, 53 pages.|
|14||Office Action dated Sep. 18, 2007 from U.S. Appl. No. 11/118,786, 22 pages.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7593904 *||Jun 30, 2005||Sep 22, 2009||Hewlett-Packard Development Company, L.P.||Effecting action to address an issue associated with a category based on information that enables ranking of categories|
|US7668789||Feb 23, 2010||Hewlett-Packard Development Company, L.P.||Comparing distributions of cases over groups of categories|
|US7797282||Sep 14, 2010||Hewlett-Packard Development Company, L.P.||System and method for modifying a training set|
|US8078604 *||Mar 19, 2007||Dec 13, 2011||Microsoft Corporation||Identifying executable scenarios in response to search queries|
|US8311957||Nov 13, 2009||Nov 13, 2012||Hewlett-Packard Development Company, L.P.||Method and system for developing a classification tool|
|US8355997||Nov 13, 2009||Jan 15, 2013||Hewlett-Packard Development Company, L.P.||Method and system for developing a classification tool|
|US8571917 *||Nov 12, 2009||Oct 29, 2013||Bank Of America Corporation||Community generated scenarios|
|US8719073||Aug 25, 2005||May 6, 2014||Hewlett-Packard Development Company, L.P.||Producing a measure regarding cases associated with an issue after one or more events have occurred|
|US9047290||Apr 29, 2005||Jun 2, 2015||Hewlett-Packard Development Company, L.P.||Computing a quantification measure associated with cases in a category|
|US9135357||Sep 20, 2010||Sep 15, 2015||Microsoft Technology Licensing, Llc||Using scenario-related information to customize user experiences|
|US9213471 *||Nov 29, 2012||Dec 15, 2015||Linkedin Corporation||Content visualization|
|US20060248054 *||Apr 29, 2005||Nov 2, 2006||Hewlett-Packard Development Company, L.P.||Providing training information for training a categorizer|
|US20070094172 *||Jul 21, 2006||Apr 26, 2007||Thaler Stephen L||Neural network based rating system|
|US20080235179 *||Mar 19, 2007||Sep 25, 2008||Microsoft Corporation||Identifying executable scenarios in response to search queries|
|US20100274830 *||Apr 23, 2009||Oct 28, 2010||Budget Tracking Solutions, Llc||Budget tracking calculator system and method|
|US20110010354 *||Sep 20, 2010||Jan 13, 2011||Microsoft Corporation||Using scenario-related information to customize user experiences|
|US20110015967 *||Jul 17, 2009||Jan 20, 2011||Gm Global Technology Operations, Inc.||Methodology to identify emerging issues based on fused severity and sensitivity of temporal trends|
|US20110112975 *||Nov 12, 2009||May 12, 2011||Bank Of America Corporation||Community generated scenarios|
|US20110119208 *||Nov 13, 2009||May 19, 2011||Kirshenbaum Evan R||Method and system for developing a classification tool|
|US20110119209 *||Nov 13, 2009||May 19, 2011||Kirshenbaum Evan R||Method and system for developing a classification tool|
|US20110119267 *||Nov 13, 2009||May 19, 2011||George Forman||Method and system for processing web activity data|
|US20110119268 *||May 19, 2011||Rajaram Shyam Sundar||Method and system for segmenting query urls|
|US20130091436 *||Nov 29, 2012||Apr 11, 2013||Linkedin Corporation||Content visualization|
|International Classification||G06E3/00, G06G7/00, G06E1/00, G06F15/18|
|Mar 21, 2006||AS||Assignment|
Owner name: HEWLETT-PACKARD DEVELOPMENT COMPANY, L.P., TEXAS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:FORMAN, GEORGE H.;SUERMONDT, HENRI J.;KIRSHENBAUM, EVAN R.;REEL/FRAME:017679/0549;SIGNING DATES FROM 20060315 TO 20060317
|May 26, 2006||AS||Assignment|
Owner name: EASTMAN CHEMICAL COMPANY, TENNESSEE
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:WOODRUFF, THOMAS EARL;WONDERS, ALAN GEORGE;REEL/FRAME:017704/0994
Effective date: 20060525
|Apr 16, 2012||FPAY||Fee payment|
Year of fee payment: 4
|Nov 9, 2015||AS||Assignment|
Owner name: HEWLETT PACKARD ENTERPRISE DEVELOPMENT LP, TEXAS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:HEWLETT-PACKARD DEVELOPMENT COMPANY, L.P.;REEL/FRAME:037079/0001
Effective date: 20151027
|Mar 24, 2016||FPAY||Fee payment|
Year of fee payment: 8